Table of content

  1. Introduction
    • Purpose
    • Objective
  2. Data
    • Description
  3. Exploratory Data Analysis
    • Mapping Police Violence
    • Police composition
    • All violent crime offenders
    • USA Election
  4. Multivariate analysis
    • All the variables
    • One-to-one analysis
    • Modeling
  5. Final analysis

1. Introduction

Purpose

Since the “Black Lives Matter” movement, the media have put in light the police violence and discrimination towards the public. Their feeling of power push them to behave like they’re above the law. In 2021, already 996 people have been killed by the police in the USA, 27% of them are Black people. Do black people have a tendency to do more crimes or is the police racist ?

We thought it would be interesting to go deeper in this analysis and discover the factors of killing rate and what influences the police to be more violent or not. We chose the United States because we knew we would find more data, moreover, having 50 states with different law policies, it is easier to compare.

In the searching of our variables of interest, we found that the location (state) factors gather interesting variables :

  • minorities percentage
  • political orientation
  • police killings rate
  • crime rate

We were motivated on this project because we think police violence is a good description of how the world is right now : separated, with still a lot of discrimination, corruption and inequalities. Every day we can see new evidence of how the police behave through social media, but can we believe everything we see ? That’s why we chose this project.

Objective

We want to observe, analyse and understand if the police is more willing to shoot a black person than a white person. We think that if a state is rather democratic or republican we can observe differences among the number of black and white people shot. We expect republican states to have more racial bias on police shooting compared to democrat ones.

Obviously, the race is also correlated to the political party of a state; white male are more willing to vote for republicans than black male.

We want to see how black and white disparities have evolved through the years and which variables influence them. Also we think that all these factors are correlated together : race , politics and police killings. Maybe we can explore other variables such as the the crime rate of a state by race, the police sex, race or age and the charges applied to the police after killing.

Is being black is a risk factor of getting shot by the police? Are republican states more willing to shoot offenders and show more racial bias? How police forces are judged regarding their position and race ? Our final objective would be to see if it is possible to predict the disparity between black and white people shot by the police according to the state demographic and politics. This will help states to put in place laws and awareness among police forces.

2. Data

Sources

For this research, we are using a variety of data sets gathered from both official and non-official sources. …

We decided to focus our analysis on a time window that goes from 2013 to 2020.

Description

For our project, we selected 4 data sets with 9 different tables from different sources : Police killings, Police officers, Presidential elections and All crime offenders. Our goal from the beginning was to split our data into states, maybe county, to really distinguish the different political party and killings rates, that’s why we were looking for complete data with the states as variables. Of course we also needed numbers for the whole territory to compare our analysis. We still think that our most significant analysis will focus at the states level because of the information we gathered, our hypothesis being that republicans states will have higher killings rate than democrats. This would be explained by the values of the political party : republicans are conservatory so we assumed they tend to have more racist or sexist tendencies. Our biggest dataset contains about 9000 values going from 2013 to 2021, however, since we decided to link this dataset to political party data we will take a time window from 2013 to 2020.

Mapping Police Violence data

Source : Mapping Police Violence

It is an unofficial website created by people to track police killings all over the United States. Their data come from different sources such as official police use of data collection programs in different states combined with nationwide data from the Fatal Encounters database. Other researches were made to complete a maximum of information : social media, obituaries, criminal records database, police reports and other.

Table 1

This database has 9678 entries of 53 variables going from 2013 to the day we uploaded it meaning the 3rd November 2021.

It contains every information on every person who was killed by the police like the date of death, their name, age, race, but also the circumstances of the murder : If the person was armed or not, if they had an illness etc. We also have information on the officer, if they were criminally charged, if they were on duty and other details.

This data set is the most interesting for our analysis because it’s the most complete. Our main goal for our project was to see if the political party of a state was correlated with the rate of police killings, meaning whereas a political party could have an impact on racism. The variables we will be using are most importantly the race, sex and state of the victim and the circumstances around their death. To do so we had to create a smaller table and clean a lot of data. From this data set we also created another table with the killings per states : killings rates and population rates.

So the initial data set was entirely cleaned and used to create additional tables. Firstly, we removed the variables that were irrelevant for us, going from 49 variables to 13. We renamed each variable in order to make them easier to use in our code. Then, we did basic modifications such as filtering the variables with NAs, changing some observations to make the understanding clearer, for instance, the variable “Fleing” had more than 10 different observation such as : car, foot, Car, other, Fleing, not fleing, etc. So in order to use these we changed the variables to yes, no and other. We applied that pattern to every other variables. This first table will help us analyze the circumstances of death and see the tendencies in order to make conclusions.
After that we subtracted a table with the numbers of killings per year for each state and population in order to have an evolution of the population and the killings.

For this we estimated average number of killings per year going from 2013 to 2020 (for each population and in total).

Table 2

The second table from the same source is a data set on the total killings per State from 2013 to 2021. It has 999 rows for 36 variables.

In this table we have information on the total population and share of ethnicity in each State but also the number of killings for each ethnicity and the disparities in between them. We wanted to analyse the average killing per year for each state and order them from the highest to the lowest. This analysis would also help us to follow the evolution of different states from 2016 to 2020 presidential elections and see if the political party had a significant impact.

For our project, we focused only on Black and White disparity since it’s our main objective. So regarding the cleaning, we first removed the NAs rows to have only the 51 States and then removed all the variables which didn’t concern Black or White population data. We were interested in comparing for each State the share of Black and White people killed compared to their population count.

Police officers data

Source : datausa.io

Datausa.io is a website of public US Government data. It was created and made by Deloitte, Datawheel and Cesar Hidalgo (MIT Media Lab professor and director of Collective Learning) in 2014. This website is a research engine for anyone to have information on cities and places regarding education like statistics and visualization on universities or best skills for a specific job. Data comes from official sources.

Table 1 : Police gender composition

The first data set we downloaded from the website is the gender composition of the US police workforce. It has 12 entries and 11 variables. Obviously, it contains information on the police officers gender rate in each state from 2014 to 2019.

We thought we would use this table to analyse the proportion of male and female in the police workforce as well as their representation. We might look for other data set concerning the proportion of male and female responsible for the police killings.

(We thought we’d use this table to see if male or female were considered more violent regarding the killings and what was their proportion in the police workforce. Did the gender played a role in these events or if it was obsolete ? This data set doesn’t really have a link with our main question research and that is why we didn’t mind the time window being different. )

Table 2 : Police ethnicity and race

This table contains 189 observations for 16 variables. It concerns the composition of ethnicity and race in the police workorce from 2014 to 2019.

Once more, does the race have an impact on the killings, meaning is the police racist ? In our hypothesis, White people kill Black people. Discrimination still exists for sure today, you can see it everywhere : at work, on the streets, on the medias,… Through our analysis we want to see the proportion of Black and White population against the proportion of Black and White police officers. Are Black people well represented in the police workforce ? Or is there discrimination at hiring procedures ?

Presidential election data

Source : worldpopultationreview.com

Table 1 : 2016 presidential elections

Our first data set for the 2016 presidential elections comes from the website worldpopulationreview.com. We could download the the data in CSV file directly from the page we were interested in. This website, as it states, review the world population and its growth. It shows information about demographics for each continents, countries, cities but also on US States and Counties. The sources used are official, coming from United Nations population.un.org data and United States Census Bureau.

This table countains 51 observations for 7 variables. It makes a census of the votes per states and which party won in 2016.

Table 2 : 2020 presidential elections

The second data set for the 2020 presidential elections comes from the website Kaggle.com. (We know it wasn’t advised to use this website, however it was the best one we found and since we’re only two on the project and already started, we decided to keep this data set). It is an online platform of data scientists and machine learning practitioners sharing their research and analysis and from worldpopulationreview.com, an independent organization without political affiliations.

The data set contains 52 observations for 6 variables. It is the same as the previous table on the 2016 elections, only it doesn’t have a “win” variable.

Our objective by using these data sets was to compare them with our police killings analysis. We wanted to do some visualizations and interpretation on the country overall : does democrat or republican dominates ? How did the elections evolved between 2016 and 2020 and did the pattern had a correlation with the evolution of the killings ?

To accomplish our goal we had to clean the tables first. We created new dataframes and rename the columns : the new table regrouped all the result of the election of 2016 and 2020 by merging the previous one on the state code. This step done, we rename the columns to focus on the percentage of democrats for each state in the USA. This focus is related to our hypothesis being that republicans tend to be more racist than democrats. Our wish is to see that a rise of democratic states is correlated to a decrease of police killings.

All offenders crime data

Source : crime-data-explorer.fr.cloud.gov

crime-data-explorer.fr.cloud.gov is an American website from the FBI’s Uniform Crime Reporting Program’s solution to have a better vision on the constant change in crime. It gathers national and state data on all crimes across the country. It is presented as a research engine in terms of States and year and can display graphs. The website allows us to view trends, download data and access Crime Data API.
We downloaded data on the offenders sex and race, giving a table of 8 observations of 6 variables. (We removed the years prior to 2013, matching our project time frame). The table display the total count on : all offenders, black and white offenders, male and female offenders.

We added this data set for our research for another hypothesis we had. In case we couldn’t find a correlation between politics and police killings, we thought crime rate could be a factor of police killings : We thought that police officers might be in more stress when they have a lot of work or face more crimes, this stress could lead to impulsive decisions or bias thinking and results in murder.Our hypothesis is, the higher is the crime rate from a certain population, the higher is the police killings towards this population. For instance : If Black people are more responsible for violent crimes it could explain why they have a higher killing rate.

For this table we wanted to extract the rate of killings per year for each state and how it evolved. If possible identify whether Black or White people were more responsible of these crimes aswell compared to their population proportion.

Crime Rate

Source : ucr.fbi.gov/Crime in the U.S.

The Uniform Crime Reporting (UCR) Program delivers accurate data for law enforcement. Students of criminal justice, scholars, the media, and the general public can all benefit from it. Since its inception in 1930, the program has provided crime data.

More than 18,000 city, university, and college law enforcement agencies, as well as county, state, tribal, and federal authorities, provide data to the UCR Program. Agencies voluntarily join and submit crime data to the FBI’s UCR Program or through a state UCR program.

We gathered data between 2010 and 2019 about the crime rate in every state in the United States. We think Crime rate is essential to explain the shooting rate in a state and create a better model.

The data set contains 52 observations for 11 variables.

For further analysis discuss later, We merge only crime rate during 2016 of this table on the other one.

3. Exploratory Data Analysis

Mapping Police Violence

Table 1 : 2013 to 2021 police killings

1) Cleaning the dataset

To get started, we had to rearrange the data set in order to retain only the variables and observations we needed. This data set was huge, we had to concentrate on the essential : Moving from 53 variables to 12. Then we did basic modifications such as changing variable names, filtering the variables with NAs, changing some observations to make the understanding clearer. We also decided to not take into account the year 2021 since it isn’t complete.

Our most important variables were :

  • Race
  • Sex
  • State
  • Date
  • Armed Status
  • Mental illness
  • Criminal charges
  • Fleeing
  • Body camera

These are the ones we are going to link with other data sets to run our hypothesis.

2) Analysing our variables

In this part we analysed our most important variables :

  • Average Black and White people killed per State and per Year
  • Percentage of black and white people killed
  • Percentage of male and female killed
  • Percentage of officers charged
  • Percentage of officers wearing a body camera
  • Percentage of victim’s armed
  • Percentage of victim’s showing a mental illness or incapacity to be clear minded
Table on black killings per state
State 2013 2014 2015 2016 2017 2018 2019 2020 Average Black killing per year
CA 30 23 36 25 29 21 25 19 26.00
FL 27 30 34 19 20 23 17 26 24.50
TX 22 27 23 21 20 21 28 21 22.88
GA 12 13 13 12 14 23 12 24 15.38
IL 13 18 13 19 15 12 9 7 13.25
LA 8 9 15 12 14 7 18 11 11.75
OH 13 13 15 12 11 9 9 11 11.62
MD 11 16 15 12 7 9 10 8 11.00
NY 13 10 11 11 8 7 12 13 10.62
NC 14 10 8 15 9 10 11 7 10.50
MO 10 12 8 8 14 9 14 8 10.38
PA 9 5 9 10 8 10 7 6 8.00
AL 13 7 7 7 8 4 9 7 7.75
VA 5 3 13 9 8 12 3 5 7.25
NJ 6 9 6 9 8 7 6 4 6.88
OK 6 7 12 7 3 6 10 4 6.88
MI 7 4 7 7 8 6 3 7 6.12
IN 11 2 7 5 7 4 4 6 5.75
TN 4 6 5 3 4 10 9 5 5.75
MS 4 8 3 5 7 5 10 3 5.62
SC 8 5 9 5 3 4 4 5 5.38
AR 6 2 1 3 5 7 8 5 4.62
AZ 6 5 1 4 7 5 3 5 4.50
WI 4 3 2 5 7 3 3 6 4.12
WA 4 2 0 4 8 0 7 5 3.75
CO 1 2 5 2 2 4 7 5 3.50
NV 1 4 3 2 4 4 3 4 3.12
DC 6 3 6 5 1 1 1 1 3.00
KY 2 3 4 2 1 3 4 2 2.62
MN 3 1 3 3 2 2 4 2 2.50
MA 5 1 2 5 1 0 2 0 2.00
KS 0 4 0 2 3 2 0 0 1.38
WV 1 1 2 1 1 3 2 0 1.38
DE 1 1 2 0 4 1 0 1 1.25
CT 2 1 0 1 1 0 2 2 1.12
NE 1 1 2 2 0 1 1 1 1.12
OR 1 0 1 1 2 3 1 0 1.12
UT 0 1 1 0 1 3 2 1 1.12
IA 0 0 0 3 1 1 1 1 0.88
NM 1 2 0 0 0 1 0 1 0.62
AK 0 1 0 0 1 1 1 0 0.50
RI 0 1 0 1 0 1 0 0 0.38
HI 0 0 1 0 0 0 0 0 0.12
ID 0 0 0 0 0 1 0 0 0.12
ME 0 0 0 0 1 0 0 0 0.12
MT 0 0 0 0 0 0 0 0 0.00

In this table we rank the States from the one with the highest count of Black killings from 2013 to 2020. We excluded 2021 since the year wasn’t over and the analysis would be incomplete. We can see that the 3 states with the highest count are : California, Florida and Texas. And the 3 states with the lowest count are : Idaho, Maine and Montana.

Table on White killings per state
State 2013 2014 2015 2016 2017 2018 2019 2020 Average White killing per year
CA 58 49 60 46 55 34 41 37 47.50
TX 38 43 48 40 27 34 37 25 36.50
FL 33 43 26 39 34 42 30 39 35.75
AZ 25 18 28 22 18 26 10 13 20.00
OK 9 14 23 18 19 23 21 15 17.75
WA 19 25 12 16 17 14 14 13 16.25
CO 11 8 11 19 17 24 15 21 15.75
TN 13 15 15 18 22 15 19 9 15.75
OH 12 15 18 13 23 22 13 9 15.62
NC 13 12 16 19 11 16 17 19 15.38
GA 12 11 20 11 20 21 18 7 15.00
MO 12 12 13 14 19 11 14 12 13.38
KY 4 15 12 17 16 15 8 14 12.62
OR 7 13 15 15 8 13 15 7 11.62
AL 5 10 11 19 17 10 4 11 10.88
IN 6 8 12 10 12 12 13 13 10.75
PA 16 7 9 11 15 14 6 7 10.62
WI 9 5 6 12 15 6 10 11 9.25
SC 4 13 11 13 7 8 12 4 9.00
NV 7 13 14 5 8 7 3 13 8.75
AR 9 4 4 13 7 13 11 7 8.50
MI 5 10 11 9 6 11 7 8 8.38
UT 6 12 10 9 3 10 6 10 8.25
NY 5 9 10 9 4 6 10 7 7.50
MS 1 7 11 5 9 6 14 5 7.25
VA 5 4 8 8 14 6 7 5 7.12
LA 4 11 11 8 4 10 2 5 6.88
MN 7 7 9 9 6 7 5 5 6.88
KS 4 9 9 9 6 4 8 5 6.75
WV 6 5 7 11 9 4 9 3 6.75
IL 9 9 6 10 6 8 2 3 6.62
NM 6 7 9 7 5 5 5 3 5.88
ID 4 2 6 4 4 11 6 3 5.00
MD 8 2 4 5 5 2 7 5 4.75
IA 3 7 5 2 4 8 4 4 4.62
MT 5 3 4 5 3 5 5 6 4.50
ME 5 6 2 1 9 2 2 4 3.88
MA 5 3 5 5 3 4 0 4 3.62
NJ 2 1 8 4 2 5 4 3 3.62
AK 1 1 1 4 4 3 4 6 3.00
NE 2 5 7 6 0 0 2 2 3.00
CT 6 0 4 4 3 1 1 1 2.50
NH 2 1 4 2 3 2 2 3 2.38
WY 1 2 4 1 1 3 0 2 1.75
SD 1 1 1 3 2 2 2 1 1.62
VT 1 0 1 1 1 2 3 1 1.25
DE 0 2 2 1 3 0 0 0 1.00
ND 1 1 0 1 0 4 0 1 1.00
HI 2 0 0 0 1 1 1 0 0.62
RI 1 0 0 0 1 0 0 0 0.25
DC 0 0 0 0 1 0 0 0 0.12

Here is the second table with the ranking of the States from the highest to the lowest average killing of White people from 2013 to 2020. We can see that the 3 first States are the same as the 1st table, however the 3 lasts are different : Hawaii, Rhode Island, District of Columbia.

Death circumstances Black people
Status Illness, alcohol or drugs Fleing Body_camera Armed Charged or sentenced
FALSE 88.24 63.88 83.91 31.75 98
TRUE 11.76 36.12 16.09 68.25 2
Death circumstances White people
Status Illness, alcohol or drugs Fleing Body_camera Armed Charged or sentenced
FALSE 76.07 72.79 90.73 26.04 99.29
TRUE 23.93 27.21 9.27 73.96 0.71

We created two tables comparing circumstances in Black and White killings. We kept the data for 2021 because we thought with more data this analysis would be more precise. We can see once again that there is a clear difference between the two races :

  • Black people are 2 times more likely to get killed regardless of their mental illness or incapacity to be clear minded
  • Black people are 1.3 times more likely to get killed if they’re fleeing
  • Black people are 1.2 times more likely to be unarmed when killed

However we can see that from 2013 to 2021 :

  • 0.71% of White killings ended up charged and convicted with a crime, against 2% for Black killings, meaning that it seems like sentence over black killings are “harder”. An hypothesis could be that the case when an officer was convicted was because the “murder” wasn’t legitimate. It would mean that there are more Black killings that tend to be unexplained.
  • 16.12% police officers were wearing a body camera when they killed a black person against 9.28% for white people. In our hypothesis it could mean that police officers are less scared about the sentence they could get when the crime concerns a black person, however as we saw above, it doesn’t match reality.

These entries don’t mean a lot of things like that, because there are other factors. It would be interesting to correlate the circumstances in between them.

To go further we wanted to see how much people were killed regardless of the circumstances meaning : Not fleeing, Unarmed and not showing mental illness (or incapacity to be clear minded)

Black people killing circumstances
race == “Black” mental_illness == “No” armed_status == “Unarmed” fleing == “no” n
FALSE FALSE FALSE FALSE 283
FALSE FALSE FALSE TRUE 1268
FALSE FALSE FALSE NA 775
FALSE FALSE TRUE FALSE 39
FALSE FALSE TRUE TRUE 198
FALSE FALSE TRUE NA 175
FALSE TRUE FALSE FALSE 1138
FALSE TRUE FALSE TRUE 1660
FALSE TRUE FALSE NA 1277
FALSE TRUE TRUE FALSE 91
FALSE TRUE TRUE TRUE 168
FALSE TRUE TRUE NA 199
TRUE FALSE FALSE FALSE 78
TRUE FALSE FALSE TRUE 254
TRUE FALSE FALSE NA 165
TRUE FALSE TRUE FALSE 21
TRUE FALSE TRUE TRUE 53
TRUE FALSE TRUE NA 66
TRUE TRUE FALSE FALSE 466
TRUE TRUE FALSE TRUE 579
TRUE TRUE FALSE NA 476
TRUE TRUE TRUE FALSE 63
TRUE TRUE TRUE TRUE 75
TRUE TRUE TRUE NA 110
NA FALSE NA NA 1
White people killing circumstances
race == “White” mental_illness == “No” armed_status == “Unarmed” fleing == “no” n
FALSE FALSE FALSE FALSE 158
FALSE FALSE FALSE TRUE 676
FALSE FALSE FALSE NA 480
FALSE FALSE TRUE FALSE 37
FALSE FALSE TRUE TRUE 114
FALSE FALSE TRUE NA 129
FALSE TRUE FALSE FALSE 985
FALSE TRUE FALSE TRUE 1297
FALSE TRUE FALSE NA 1137
FALSE TRUE TRUE FALSE 111
FALSE TRUE TRUE TRUE 145
FALSE TRUE TRUE NA 196
TRUE FALSE FALSE FALSE 203
TRUE FALSE FALSE TRUE 846
TRUE FALSE FALSE NA 460
TRUE FALSE TRUE FALSE 23
TRUE FALSE TRUE TRUE 137
TRUE FALSE TRUE NA 112
TRUE TRUE FALSE FALSE 619
TRUE TRUE FALSE TRUE 942
TRUE TRUE FALSE NA 616
TRUE TRUE TRUE FALSE 43
TRUE TRUE TRUE TRUE 98
TRUE TRUE TRUE NA 113
NA FALSE NA NA 1

We want to focus on the lines where we have on the first table :

  • FALSE, TRUE, TRUE, TRUE : This represent people who aren’t Black who were killed despite showing no apparent reason to be. As a result, 168 people were killed in these circumstances.

On both tables :

  • TRUE, TRUE, TRUE, TRUE :
  • All Black people who were killed but showing no apparent reason to be. As a result, 75 Black people were killed in these circumstances. It means 30,86% of the total people killed in this situation.
  • All White people who were killed but showing no apparent reason to be. As a result, 98 White people were killed in these circumstances. It means 40,32% of the total people killed in this situation.
Total people killed from 2013 to 2020
year white people killed perc_wpk black people killed perc_bpk other people killed perc_opk total people killed
2013 430 39.56 291 26.77 366 33.67 1087
2014 480 45.76 276 26.31 293 27.93 1049
2015 543 49.27 305 27.68 254 23.05 1102
2016 533 49.81 279 26.07 258 24.11 1070
2017 509 46.61 278 25.46 305 27.93 1092
2018 512 44.76 266 23.25 366 31.99 1144
2019 449 40.97 282 25.73 365 33.30 1096
2020 411 36.47 249 22.09 467 41.44 1127
Summary of the total killings from 2013 to 2020
year white people killed perc_wpk black people killed perc_bpk other people killed perc_opk total people killed
Length:8 Min. :411.0 Min. :36.47 Min. :249.0 Min. :22.09 Min. :254.0 Min. :23.05 Min. :1049
Class :character 1st Qu.:444.2 1st Qu.:40.62 1st Qu.:273.5 1st Qu.:24.91 1st Qu.:284.2 1st Qu.:26.98 1st Qu.:1083
Mode :character Median :494.5 Median :45.26 Median :278.5 Median :25.90 Median :335.0 Median :29.96 Median :1094
NA Mean :483.4 Mean :44.15 Mean :278.2 Mean :25.42 Mean :334.2 Mean :30.43 Mean :1096
NA 3rd Qu.:517.2 3rd Qu.:47.27 3rd Qu.:284.2 3rd Qu.:26.43 3rd Qu.:366.0 3rd Qu.:33.39 3rd Qu.:1108
NA Max. :543.0 Max. :49.81 Max. :305.0 Max. :27.68 Max. :467.0 Max. :41.44 Max. :1144

In average, 1096 people were killed each year since 2013, 25,42% being Black people and 44,15% being White people.

3) Graphic vizualisation

## <!-- BarChart generated in R 4.1.1 by googleVis 0.6.10 package -->
## <!-- Sat Dec 18 12:47:46 2021 -->
## 
## 
## <!-- jsHeader -->
## <script type="text/javascript">
##  
## // jsData 
## function gvisDataBarChartID2cee75037236 () {
## var data = new google.visualization.DataTable();
## var datajson =
## [
##  [
## "CA",
## 30,
## 23,
## 36,
## 25,
## 29,
## 21,
## 25,
## 19,
## 26
## ],
## [
## "FL",
## 27,
## 30,
## 34,
## 19,
## 20,
## 23,
## 17,
## 26,
## 24.5
## ],
## [
## "TX",
## 22,
## 27,
## 23,
## 21,
## 20,
## 21,
## 28,
## 21,
## 22.88
## ],
## [
## "GA",
## 12,
## 13,
## 13,
## 12,
## 14,
## 23,
## 12,
## 24,
## 15.38
## ],
## [
## "IL",
## 13,
## 18,
## 13,
## 19,
## 15,
## 12,
## 9,
## 7,
## 13.25
## ],
## [
## "LA",
## 8,
## 9,
## 15,
## 12,
## 14,
## 7,
## 18,
## 11,
## 11.75
## ],
## [
## "OH",
## 13,
## 13,
## 15,
## 12,
## 11,
## 9,
## 9,
## 11,
## 11.62
## ],
## [
## "MD",
## 11,
## 16,
## 15,
## 12,
## 7,
## 9,
## 10,
## 8,
## 11
## ],
## [
## "NY",
## 13,
## 10,
## 11,
## 11,
## 8,
## 7,
## 12,
## 13,
## 10.62
## ],
## [
## "NC",
## 14,
## 10,
## 8,
## 15,
## 9,
## 10,
## 11,
## 7,
## 10.5
## ],
## [
## "MO",
## 10,
## 12,
## 8,
## 8,
## 14,
## 9,
## 14,
## 8,
## 10.38
## ],
## [
## "PA",
## 9,
## 5,
## 9,
## 10,
## 8,
## 10,
## 7,
## 6,
## 8
## ],
## [
## "AL",
## 13,
## 7,
## 7,
## 7,
## 8,
## 4,
## 9,
## 7,
## 7.75
## ],
## [
## "VA",
## 5,
## 3,
## 13,
## 9,
## 8,
## 12,
## 3,
## 5,
## 7.25
## ],
## [
## "NJ",
## 6,
## 9,
## 6,
## 9,
## 8,
## 7,
## 6,
## 4,
## 6.88
## ],
## [
## "OK",
## 6,
## 7,
## 12,
## 7,
## 3,
## 6,
## 10,
## 4,
## 6.88
## ],
## [
## "MI",
## 7,
## 4,
## 7,
## 7,
## 8,
## 6,
## 3,
## 7,
## 6.12
## ],
## [
## "IN",
## 11,
## 2,
## 7,
## 5,
## 7,
## 4,
## 4,
## 6,
## 5.75
## ],
## [
## "TN",
## 4,
## 6,
## 5,
## 3,
## 4,
## 10,
## 9,
## 5,
## 5.75
## ],
## [
## "MS",
## 4,
## 8,
## 3,
## 5,
## 7,
## 5,
## 10,
## 3,
## 5.62
## ],
## [
## "SC",
## 8,
## 5,
## 9,
## 5,
## 3,
## 4,
## 4,
## 5,
## 5.38
## ],
## [
## "AR",
## 6,
## 2,
## 1,
## 3,
## 5,
## 7,
## 8,
## 5,
## 4.62
## ],
## [
## "AZ",
## 6,
## 5,
## 1,
## 4,
## 7,
## 5,
## 3,
## 5,
## 4.5
## ],
## [
## "WI",
## 4,
## 3,
## 2,
## 5,
## 7,
## 3,
## 3,
## 6,
## 4.12
## ],
## [
## "WA",
## 4,
## 2,
## 0,
## 4,
## 8,
## 0,
## 7,
## 5,
## 3.75
## ],
## [
## "CO",
## 1,
## 2,
## 5,
## 2,
## 2,
## 4,
## 7,
## 5,
## 3.5
## ],
## [
## "NV",
## 1,
## 4,
## 3,
## 2,
## 4,
## 4,
## 3,
## 4,
## 3.12
## ],
## [
## "DC",
## 6,
## 3,
## 6,
## 5,
## 1,
## 1,
## 1,
## 1,
## 3
## ],
## [
## "KY",
## 2,
## 3,
## 4,
## 2,
## 1,
## 3,
## 4,
## 2,
## 2.62
## ],
## [
## "MN",
## 3,
## 1,
## 3,
## 3,
## 2,
## 2,
## 4,
## 2,
## 2.5
## ],
## [
## "MA",
## 5,
## 1,
## 2,
## 5,
## 1,
## 0,
## 2,
## 0,
## 2
## ],
## [
## "KS",
## 0,
## 4,
## 0,
## 2,
## 3,
## 2,
## 0,
## 0,
## 1.38
## ],
## [
## "WV",
## 1,
## 1,
## 2,
## 1,
## 1,
## 3,
## 2,
## 0,
## 1.38
## ],
## [
## "DE",
## 1,
## 1,
## 2,
## 0,
## 4,
## 1,
## 0,
## 1,
## 1.25
## ],
## [
## "CT",
## 2,
## 1,
## 0,
## 1,
## 1,
## 0,
## 2,
## 2,
## 1.12
## ],
## [
## "NE",
## 1,
## 1,
## 2,
## 2,
## 0,
## 1,
## 1,
## 1,
## 1.12
## ],
## [
## "OR",
## 1,
## 0,
## 1,
## 1,
## 2,
## 3,
## 1,
## 0,
## 1.12
## ],
## [
## "UT",
## 0,
## 1,
## 1,
## 0,
## 1,
## 3,
## 2,
## 1,
## 1.12
## ],
## [
## "IA",
## 0,
## 0,
## 0,
## 3,
## 1,
## 1,
## 1,
## 1,
## 0.88
## ],
## [
## "NM",
## 1,
## 2,
## 0,
## 0,
## 0,
## 1,
## 0,
## 1,
## 0.62
## ],
## [
## "AK",
## 0,
## 1,
## 0,
## 0,
## 1,
## 1,
## 1,
## 0,
## 0.5
## ],
## [
## "RI",
## 0,
## 1,
## 0,
## 1,
## 0,
## 1,
## 0,
## 0,
## 0.38
## ],
## [
## "HI",
## 0,
## 0,
## 1,
## 0,
## 0,
## 0,
## 0,
## 0,
## 0.12
## ],
## [
## "ID",
## 0,
## 0,
## 0,
## 0,
## 0,
## 1,
## 0,
## 0,
## 0.12
## ],
## [
## "ME",
## 0,
## 0,
## 0,
## 0,
## 1,
## 0,
## 0,
## 0,
## 0.12
## ],
## [
## "MT",
## 0,
## 0,
## 0,
## 0,
## 0,
## 0,
## 0,
## 0,
## 0
## ] 
## ];
## data.addColumn('string','State');
## data.addColumn('number','2013');
## data.addColumn('number','2014');
## data.addColumn('number','2015');
## data.addColumn('number','2016');
## data.addColumn('number','2017');
## data.addColumn('number','2018');
## data.addColumn('number','2019');
## data.addColumn('number','2020');
## data.addColumn('number','Average Black killing per year');
## data.addRows(datajson);
## return(data);
## }
##  
## // jsDrawChart
## function drawChartBarChartID2cee75037236() {
## var data = gvisDataBarChartID2cee75037236();
## var options = {};
## options["allowHtml"] = true;
## options["title"] = "Interactive Bar chart";
## options["legend"] = "right";
## options["pointSize"] = 10;
## 
## 
##     var chart = new google.visualization.BarChart(
##     document.getElementById('BarChartID2cee75037236')
##     );
##     chart.draw(data,options);
##     
## 
## }
##   
##  
## // jsDisplayChart
## (function() {
## var pkgs = window.__gvisPackages = window.__gvisPackages || [];
## var callbacks = window.__gvisCallbacks = window.__gvisCallbacks || [];
## var chartid = "corechart";
##   
## // Manually see if chartid is in pkgs (not all browsers support Array.indexOf)
## var i, newPackage = true;
## for (i = 0; newPackage && i < pkgs.length; i++) {
## if (pkgs[i] === chartid)
## newPackage = false;
## }
## if (newPackage)
##   pkgs.push(chartid);
##   
## // Add the drawChart function to the global list of callbacks
## callbacks.push(drawChartBarChartID2cee75037236);
## })();
## function displayChartBarChartID2cee75037236() {
##   var pkgs = window.__gvisPackages = window.__gvisPackages || [];
##   var callbacks = window.__gvisCallbacks = window.__gvisCallbacks || [];
##   window.clearTimeout(window.__gvisLoad);
##   // The timeout is set to 100 because otherwise the container div we are
##   // targeting might not be part of the document yet
##   window.__gvisLoad = setTimeout(function() {
##   var pkgCount = pkgs.length;
##   google.load("visualization", "1", { packages:pkgs, callback: function() {
##   if (pkgCount != pkgs.length) {
##   // Race condition where another setTimeout call snuck in after us; if
##   // that call added a package, we must not shift its callback
##   return;
## }
## while (callbacks.length > 0)
## callbacks.shift()();
## } });
## }, 100);
## }
##  
## // jsFooter
## </script>
##  
## <!-- jsChart -->  
## <script type="text/javascript" src="https://www.google.com/jsapi?callback=displayChartBarChartID2cee75037236"></script>
##  
## <!-- divChart -->
##   
## <div id="BarChartID2cee75037236" 
##   style="width: 500; height: automatic;">
## </div>

Here is a graphic representation of the Black killings table showing the average number of black people killed in each state from 2013 to 2020.

We can see that between 2013 and 2020, White people were the most killed and Black people are the second most killed population. The proportion of women killed is very little compared to men. However we can observe that White, Black and Hispanic women are the most affected. Since women represent such a little proportion of the killings (5%) we won’t be analyzing further on gender.

Here is a representation of the circumstances of death comparing Black and White people when killed.

Killings per year

The graph shows globally that the killings dropped a little since 2018, a decrease of about 150. We can still say that Black and White killings decreased since 2015.

We wanted to show the evolution of black killings in the 3 most violent states. We can observe that in 2016 there’s a big decrease in all states, 2016 being the Trump presidential elections. However Trump didn’t start the service until January 2017 meaning that Barack Obama was still president at that time. We know that in 2015 the Black Lives Matter movement started to spread and people asked political personalities to take action. It could explain the decrease of killings in 2016 in these states.

Table 2 : 2013 to 2021 State killings

1) Cleaning the data set

The second table is a summary of the killings by state. We have a table of 999 observations of 36 variables. We already see a problem because we should have 50 observations corresponding to the states.
In this part of the analysis, the data goes from 2013 to 2021 and we will keep this time frame because it represents the total killings.

2) Analyzing our variables

We want to focus only on black versus white killings. Our main variables will be :

  • State
  • State abbreviation
  • Total population
  • Black population
  • White population
  • Black people killed
  • White people killed

Our goal is to compare the ratio between Black and White killings to their population in each State. So we created a new data frame subtracted from the original one. We added columns for the White population killed and its percentage.

State State Abbreviation Total Population % Black % White # People Killed % Black victims % White victims
California CA 39148760 5.53 37.54 1348 15.13 27.30
Texas TX 27885195 11.72 42.34 804 22.26 34.95
Florida FL 20598139 15.38 54.36 638 30.56 44.20
New York NY 19618453 14.32 55.86 179 46.37 33.52
Illinois IL 12821497 14.01 61.57 195 53.85 27.18
Pennsylvania PA 12791181 10.64 76.83 192 33.33 44.27
Ohio OH 11641879 12.18 79.20 237 38.82 52.74
Georgia GA 10297484 31.03 53.18 309 39.16 38.83
North Carolina NC 10155624 21.13 63.34 239 35.15 51.46
Michigan MI 9957488 13.66 75.21 139 34.53 46.04
New Jersey NJ 8881845 12.71 55.84 113 48.67 25.66
Virginia VA 8413774 18.81 62.20 140 41.43 40.00
Washington WA 7294336 3.56 69.08 253 11.86 50.59
Arizona AZ 6946685 4.13 55.07 392 8.42 40.56
Massachusetts MA 6830193 6.79 72.19 61 24.59 47.54
Tennessee TN 6651089 16.65 74.04 207 22.22 60.87
Indiana IN 6637426 9.20 79.48 152 30.26 55.92
Missouri MO 6090062 11.49 79.61 230 35.22 46.09
Maryland MD 6003435 29.31 51.39 144 60.42 26.39
Wisconsin WI 5778394 6.26 81.53 130 24.62 53.85
Colorado CO 5531141 3.92 68.31 272 10.29 45.22
Minnesota MN 5527358 6.09 80.29 94 21.28 58.51
South Carolina SC 4955925 26.80 63.70 134 32.09 53.73
Alabama AL 4864680 26.43 65.71 158 37.34 53.16
Louisiana LA 4663616 32.00 58.84 173 53.76 31.79
Kentucky KY 4440204 7.87 84.77 140 15.00 70.71
Oregon OR 4081943 1.82 76.03 127 7.09 73.23
Oklahoma OK 3918137 7.21 66.00 245 22.45 57.55
Connecticut CT 3581504 9.82 67.53 41 21.95 48.78
Iowa IA 3132499 3.43 86.06 51 13.73 72.55
Utah UT 3045350 1.11 78.62 98 9.18 65.31
Arkansas AR 2990671 15.33 72.69 117 29.91 57.26
Mississippi MS 2988762 37.53 56.78 120 37.50 48.33
Nevada NV 2922849 8.55 49.89 147 17.01 46.94
Kansas KS 2908776 5.63 76.13 81 13.58 66.67
New Mexico NM 2092434 1.82 37.67 164 3.05 28.05
Nebraska NE 1904760 4.64 79.40 40 22.50 60.00
West Virginia WV 1829054 3.59 92.11 79 13.92 65.82
Idaho ID 1687809 0.64 82.22 57 1.75 68.42
Hawaii HI 1422029 1.72 22.12 40 2.50 12.50
New Hampshire NH 1343622 1.31 90.39 21 0.00 85.71
Maine ME 1332813 1.27 93.38 35 2.86 88.57
Rhode Island RI 1056611 5.58 72.66 6 50.00 33.33
Montana MT 1041732 0.42 86.29 51 0.00 70.59
Delaware DE 949495 21.56 62.73 22 45.45 36.36
South Dakota SD 864289 1.83 82.24 27 0.00 44.44
North Dakota ND 752201 2.67 85.01 14 0.00 57.14
Alaska AK 738516 3.09 61.04 49 8.16 48.98
District of Columbia DC 684498 46.06 36.24 27 88.89 3.70
Vermont VT 624977 1.22 92.97 13 0.00 76.92
Wyoming WY 581836 0.88 84.14 23 0.00 60.87

We can observe in this table that the percent of Black people killed compared to their population rate is higher than the White killings. It means that even though Black people represent less in each State, they’re still more killed than White people. Let’s take California for instance : Black people represent only 5.53% of the population but 15% of people killed are Black. Whereas White people represent 37.54% and only 27.3%of people killed are White.

Summary of Black and White people killed
Race Total_pop proportion_tp pop_killed proportion_k
Total 322903030 100.00 8767 0.003
Black 39715917 12.30 2226 0.006
White 197181177 61.07 3867 0.002

After some calculation, we fount out that from 2013 to 2020, black people were killed 3 times more than white people : 0.006% of black people are killed against 0.002% for white people. This clearly means that there is an existing discrimination against black people. However, it could be explained by other factors like the crime rate, and the circumstances of death, still there is a significant difference.

3) Graphic vizualisation

This graph compares the percentage of black people killed with the percentage of black population in each state. The black points representing the percentage of Black people killed are clearly above the red points meaning that the proportion of Black people killed is higher than their population proportion in almost every State. For instance in DC (District of Columbia), Black people represent 46% of the population but 89% of the total police killings are Black people. Drawing a conclusion on this graph we could say that there is clearly a tendency from the police to kill Black people more.

Police composition

Table 1 : Gender composition

1) Cleaning the data set

Gender Year share
Male 2019 84.90
Female 2019 15.10
Male 2018 85.11
Female 2018 14.89
Male 2017 85.94
Female 2017 14.06
Male 2016 86.15
Female 2016 13.85
Male 2015 85.80
Female 2015 14.20
Male 2014 85.65
Female 2014 14.35

We only kept this 3 variables to have a clear vision of the evolution among time of the gender diversity in the police.

2) Data vizualisation

In the United States in 2020, there is roughly 51% of women and the Black population represents approximately 13,5% of the total population.

On the opposite, the number of women in the police is really low. There has always been more women than men in the United States but still close to 50&%. Indeed, women represent about 51% against 49% of men. On the graph above we clearly see that women are only 15% of the total police workforce, still in 2020.

Table 2 : Race and ethnicity composition

1) Cleaning the data set

Race PUMS.Ethnicity.Parent Year share
2 White Not Hispanic 2019 61.35
4 Black Not Hispanic 2019 11.89
20 White Not Hispanic 2018 61.95
22 Black Not Hispanic 2018 11.75
38 White Not Hispanic 2017 62.53
40 Black Not Hispanic 2017 11.62
56 White Not Hispanic 2016 63.28
58 Black Not Hispanic 2016 11.62
74 White Not Hispanic 2015 63.77
76 Black Not Hispanic 2015 11.50
92 White Not Hispanic 2014 64.54
94 Black Not Hispanic 2014 11.28

2) Data vizualisation

For the past 20 years, the percentage of black population is rising in the united state and this trend is even stronger in the force police. However, the share in the police is approximately 1.5 point down compared to the share in the united state. Though, We can still agree that the black population is correctly represent in the police.

All violent crime offenders

1) Cleaning and creating a new table

year offender_total offender_white offender_black offender_male offender_female
2020 615989 44.42 43.87 76.81 17.56
2019 528322 44.77 44.44 78.35 17.19
2018 437763 46.30 43.15 78.54 17.28
2017 387940 46.68 43.65 78.86 16.76
2016 377341 47.01 44.14 53.17 16.47
2015 357324 45.69 45.29 79.70 16.21
2014 340481 45.05 46.39 79.94 16.23
2013 339777 44.29 47.47 80.47 15.87
year offender_total offender_white offender_black offender_male offender_female
Min. :2013 Min. :339777 Min. :44.29 Min. :43.15 Min. :53.17 Min. :15.87
1st Qu.:2015 1st Qu.:353113 1st Qu.:44.68 1st Qu.:43.81 1st Qu.:77.97 1st Qu.:16.23
Median :2016 Median :382640 Median :45.37 Median :44.29 Median :78.70 Median :16.61
Mean :2016 Mean :423117 Mean :45.53 Mean :44.80 Mean :75.73 Mean :16.70
3rd Qu.:2018 3rd Qu.:460403 3rd Qu.:46.40 3rd Qu.:45.56 3rd Qu.:79.76 3rd Qu.:17.21
Max. :2020 Max. :615989 Max. :47.01 Max. :47.47 Max. :80.47 Max. :17.56

In 2020, there were 615’989 violent-crime incident compared to only 339’777 in 2013. the percentage of black offender and white offender is very close in many states.

Violence is rising in the United Sates, but what population is more responsible of it ? Let’s take a deeper look.

2) Data vizualisation

First of all, despite represent only 13.5 percent of the population, black people represent more than 40 percent of all violent crime in the United States. However, for the past years, the share of black people involved in violent crime is going down except in 2019.

The share that black people represent in the violent crime could explain why they are over killed compared to their population. In 2016, black people represent 13.3% of the population but 44.2% of violent crime. The ratio is 3.3. On the other side, they represent approximately 25% of people getting shot every year. The ratio in this the case is 1.9.

On the opposite, the percentage violent crime done by women is very low. Still, their proportion has increased by more than 1.5% since 2013.

New tables : Massachusetts and Alabama

1) Cleaning our tables

For further analysis on a state level, we decided to look for information on crime and offenders in Alabama and Massachusetts. We chose these 2 states because they differ in our interested variables : Alabama is mainly republican, and 26% of its population is black whereas Massachusetts has only 7% of Black people and is a democrat state.

We were only interested to compare the race variables, we rearrange one column to use it for our analysis. The main table had “Race_ID” and another table had the matching race to the ID, so we had to rename the observations in order to have understandable results.

Massachussetts offenders race proportion
Race Proportion
Unknown 14.54
White 60.45
Black 23.65
American indian or Alaska native 0.08
Asian 1.28
Pacific Islander 0.01
Alabama offenders race proportion
Race Proportion
Unknown 12.75
White 53.82
Black 33.03
American indian or Alaska native 0.17
Asian 0.13
Pacific Islander 0.10

To summarize, we created a table for each state that breaks down the proportion of the different races in the offender category. This helped then plotting the following graph.

2) Data vizualisation

In order to compare the states we merged variables from the Killings by State and the Crime Offenders tables. Our objective was to compare the proportion of Black people, Black offenders and Black people killed by the police.

We can observe that both in Alabama and Massachusetts the rate of Black offenders is way above the population rate. The pattern is quite similar but surprisingly, the difference between crime rate and Black victims in Massachusetts is higher than in Alabama, when the first one is majorly democrat and the second one republican.

USA Elections

1) Cleaning and creating a new table

We create a new table that regroup all the result of the election of 2016 and 2020 by merging the previous one on the state code. This step done, we rename the columns to focus on the percentage of democrats in each state in the USA.

state percent_democrat_2016 percent_democrat_2020
Alabama 34.4 37.0
Alaska 36.5 36.1
Arizona 45.1 50.3
Arkansas 33.7 35.6
California 61.7 65.9
Colorado 48.2 56.8
Connecticut 54.6 60.2
Delaware 53.1 59.6
Florida 47.8 48.3
Georgia 45.6 50.1
Hawaii 62.2 65.0
Idaho 27.5 34.1
Illinois 55.8 57.9
Indiana 37.9 41.8
Iowa 41.7 45.8
Kansas 36.0 42.3
Kentucky 32.7 36.7
Louisiana 38.5 40.5
Maine 47.8 55.2
Maryland 60.3 64.9
Massachusetts 60.0 66.8
Michigan 47.3 51.4
Minnesota 46.4 53.6
Mississippi 40.1 39.5
Missouri 38.1 42.1
Montana 35.8 41.6
Nebraska 33.7 40.1
Nevada 47.9 51.3
New Hampshire 47.0 53.6
New Jersey 55.0 59.2
New Mexico 48.3 55.4
New York 59.0 56.5
North Carolina 46.2 49.3
North Dakota 27.2 32.8
Ohio 43.6 45.9
Oklahoma 28.9 33.1
Oregon 50.1 58.3
Pennsylvania 47.5 50.3
Rhode Island 54.4 60.3
South Carolina 40.7 44.1
South Dakota 31.7 36.6
Tennessee 34.7 38.1
Texas 43.2 47.0
Utah 27.5 39.2
Vermont 56.7 67.2
Virginia 49.7 54.9
Washington 52.5 60.3
West Virginia 26.4 30.1
Wisconsin 46.5 50.3
Wyoming 21.6 27.5

Almost every state got more democrat vote in 2020 than in 2016. This led to the election of a democrat president in 2020.

2) Election map

The states that are colored in blue had 50% or more for the democratic party. We can see that overall the country presents more states that are republican (Democratic party <50%) but, tends to be more democrat in 2020. We can observe this small difference on the 2 maps.

Another detail we can add is that states that are democratic are bigger and correspond to places where there are more metropolis, business and so, more foreigners : California or New York for instance.

3) Graphic visualization

In the election of 2020, there is more state above the 50 percent line than 4 years ago. With this graph, the difference between 2016 and 2020 is easy to spot and more state are democrats in 2020.

Crime Rate

The crime rate exploded from 2014 to 2016. Since then it’s been decreasing each year, but is still higher than in 2014. 2016 corresponds to an election year : historically speaking, Trump was elected after Obama, maybe this could be a factor in the crime rate. Comparing this to the police killings rate, there’s no correlation since we saw that the killings rose from 2016 to 2018.

4. Modelling

First of all, we will build different model based on the population sample in every state of 2018, the election result of 2016, and the people shoot from 2013 to 2020.

The election result of 2016 is right in the middle of the period took for people shoot and we assume the population do not vary a lot between 2016 and 2018.

Variables

Dependant variable

Rate..All.People. : Rate of peaople killed by 125’000 inhabitants (2013-2020)

Black.White.Disparity : Disparity in rate between black and white (2013-2018)

Independent Variables

percent_democrat_2016 : percent of democrat for 2016

percent_black : percent of black in 2018

percent_asian : percent of asian in 2018

percent_white : percent of white in 2018

percent_hispanic : percent of hispanic in 2018

violent_crime_2016 : Violent crime rates per 100000 inhabitants in 2016

Total.Population : Total population in 2018

Correlation table

Rate all people is strongly correlated with violent crime 2016. This does make sense, more violent states lead to more police shooting. It is also negatively correlated with percent democrat 2016 while percent_democrat_2016 is slightly negatively correlated with violent_crime_2016.

Black white disparity is strongly negatively correlated with violent crime in 2016. On the other side, percent democrat 2016 is positevely correlated to black and white disparity. The other variable have a small influence on the black white disparity

Percent of white is srtrongly negatively correlated to percent Democrat 2016 while percent black is positevely correlated. Black population vote more for a democrat president than the white population who prefer a republican president.

Can we predict the number of people getting shot by the police with the democratic percentage?

Boxplot

Despite 2 outliers, we decided to keep both of them.

Plot 1

First of all, there is a clear negative correlation between the percentage of democrat in a stat and the number of people killed by 125’000 people. Police in more democratic state seems to kill less people every year.

Plot 2

Secondly, Violent Crime Rate is strongly correlated to the number of people killed. Obviously, more violent crimes in a state lead to more shooting by the police.

Multilinear regression

  Rate..All.People.
Predictors Estimates CI p
(Intercept) 25.81 13.87 – 37.74 <0.001
percent democrat 2016 -0.05 -0.10 – -0.01 0.020
violent crime 2016 0.00 -0.00 – 0.01 0.077
percent black -0.20 -0.32 – -0.09 0.001
percent white -0.22 -0.35 – -0.10 0.001
percent hispanic -0.13 -0.26 – -0.00 0.046
percent asian -0.35 -0.58 – -0.12 0.003
Total Population -0.00 -0.00 – 0.00 0.209
Observations 50
R2 / R2 adjusted 0.709 / 0.661

An R-squared of 0.709 is very high as well as the adjusted R-squared.

Percent white and percent black are extremely significative (p<0.001).Percent asian is also significative with 2 stars (p<0.01) and in the end percent hispanic but with only 1 star (p<0.5).

Weirdly, all demographic variable are negative.

The variable we were interested primarily, percent democrat 2016, is also significative but with p<0.05. In this regression, police in less democratic state definitely shoot more people during arrest than more democratic state.

As expected, the Total Population variable has no influence on Rate All People. Maybe we can remove this variable to get a better regression.

Regression Diagnosis

Residual vs Fitted : The curve is flat meaning there is no heteroscedasticity

Normal Q-Q : The linearity regression fit perfectly all the variable. The curve stay flat on both extreme.

Scale-Location : The red line is roughly horizontal across the plot; homoscedasticity is satisfied.

Residual vs Leverage : Observation 5 is slightly out of the red line meaning this observation has a huge impact on our regression. Globally, there is no big anomaly.

The 4 plots indicate our multilinear regression is really good and there is no issue with it.

Forward selection

All indicator suggest for the forward selection the model with all variables included (7 variables)

In the forward selection, percent_asian is surprisingly the last variable to be added to the regression and violent_crime_2016 is the first one as the plot suggested it.

Backward selection

However, for the backward selection, it is less explicit. The best model seems to be the one with 6 variables.

Obviously, Total.Population is the first variable to me removed from the backward selection process. This variable is the less significative in the base model with the 7 variables.

Final model

  Rate..All.People.
Predictors Estimates CI p
(Intercept) 28.72 17.64 – 39.80 <0.001
percent democrat 2016 -0.05 -0.10 – -0.01 0.023
violent crime 2016 0.00 -0.00 – 0.01 0.097
percent black -0.24 -0.34 – -0.14 <0.001
percent white -0.25 -0.37 – -0.14 <0.001
percent hispanic -0.17 -0.28 – -0.06 0.003
percent asian -0.41 -0.62 – -0.19 <0.001
Observations 50
R2 / R2 adjusted 0.698 / 0.656

Conclusion

To conclude, the best model is the one with all variables except Total.Population. Despite the adjusted R-squared being slightly higher with this variable, the other test proved the model is better without it.

To the question can we predict the number of people getting shot by the police with the democratic percentage? The answer is yes. We can explain 70% of the regression and most of our variable are significative.

Democratic state have less people shoot by the police. Is it because people in democratic state own less guns than republican? Is it because justice is more rehabilitative? In a further analysis, this variables could be added and we could take special measure to diminish people killed every year in state.

Can we explain the disparity between black and white shots with the demographic and democratic percentage of the state?

Boxplot

First of all, I removed 1 outliers (Black White Disprity = 19) of the model but I kept the second outliers (Black White Disprity = 10)

Plot 1

On average, the number of black people shoot by police per 125’000 inhabitant is 3.7 time higher than for white people.

This black white disparity is positively correlated to the percent of democrat in 2016.

Democrat state seems to killed more black than white per 125^000 inhabitant.

Plot 2

Moreover, violent crime rate is slightly negatively correlated to the black and white disparity.

Plot 3

As before, the line is pretty flat but still slightly negatively correlated.

In this plot, if the percentage of black people in a state is lower, the disparity between them and white people will be lower.

Multilinear regression

  Black.White.Disparity
Predictors Estimates CI p
(Intercept) -20.90 -44.94 – 3.14 0.087
percent democrat 2016 0.00 -0.09 – 0.09 0.941
violent crime 2016 0.00 -0.00 – 0.01 0.384
percent black 0.18 -0.05 – 0.41 0.116
percent white 0.24 -0.01 – 0.49 0.056
percent hispanic 0.23 -0.03 – 0.49 0.082
percent asian 0.43 -0.04 – 0.89 0.070
Total Population 0.00 -0.00 – 0.00 0.859
Observations 49
R2 / R2 adjusted 0.151 / 0.006

An R-squared of 0.15 is very low and an adjusted R-squared close to be negative is really bad.

On top of that, none of the variable are significative despite some being close to.

Basically, this multilinear regression do not predict or explain the black and white disparity and there is no need to pick a “better” model among this one.

We can still show the total population do not have an effect on the regression again as well percent democrat 2016 and violent crime 2016. At least, this might show us the disparity between black and white is not due to this variable at all.

Regression Diagnosis

Residual vs Fitted : The curve is not flat and have a weird shape. This show us there is no homoscedasticity.

Normal Q-Q : The linearity regression do not fit observation 44 that is clearly way above the line.

Scale-Location : The red line has a banana shape confirming what said above: there is heteroscedasticity.

Residual vs Leverage : Observation 11 is out of the red ine and observation 2 is close to it. This regression suffer from variable 11 that have a huge influence on it.

The 4 plots indicate our multilinear regression is not a good fit at all.

Conclusion

Despite being the important question of our research, we are unable with these data to construct a good model. This teach us at least the disparity between black and white is way more complex to explain than with only few variables.

Maybe, quadratic model would work better but there is a risk to overfit the data and it become harder to interpret the model specially with cubic model or higher.

5. Discussion

Our objective was to answer the questions :

  • Is there a real disparity between black and white people getting shot every year ?

  • Can we explain the disparity between black and white killed with the democratic percentage within a state?

  • Is race an important factor in the rate of police killings ?

  • Is the rate of crime positively correlated with the police killings regarding a certain race ? ie : If black people are more responsible of crime rate, does it mean they are killed more ?

Here is what we found out :

  • The United States represents 2 main population : White at 76% and Black at 13%.
  • Since 2013, an average number of 1096 people are killed each year. White people represent 44% of these killings and 25% are Black people, meaning that Black people were killed 3 times more than White people. Most of these people killed were male (95%). This first analysis answered our first question : yes, there is a disparity among Black and White people killed in the United States.

So how could we explain this disparity in Black and White killings ?

  • United states is a country that has 2 main party : Democratic that claim to be liberal and progressive, and Republican, conservative and traditionnal. Since a long time most president have been democratic. Our hypotethis was that the states where there were more republican vote, there would be higher rate of black killings. After analyzing our variables with the help of graphic vizualisations we couldn’t find a clear answer. For instance in Rhode Island since 2016 more than 50% of the votes were democrats. However 50% of people killed by the police were Black while representing only 5,6% of the population. Our modelling part revealed that there was no real correlation between the political party of a state and the black killings (A COMPLETER PAR JEREMY). While some states fitted the hypothesis, others didn’t, making the political party a minor factor of disparity in Black and White killings.

  • Another question we had was, if the political party didn’t explain the disparity than, did crime rate ?

In our analysis we observed that men were responsible of more than 80% of crime, while women represented approximately 16%. However, 95% of killed people were men, meaning there is even a disparity between gender. This could be explained by factors such as : the fact that women were in general less tempted to flee or to be armed, but also the fact that, and without research, in our society male are more prone to violence. Moreover, it is “inculqué” in every children that “a boy shouldn’t hit a woman”, making the man the “strong” one and the woman the one to defend. In consequence, it means that no matter the circumstance, for a lot of people it would be harder to shoot or kill a woman.

Regarding crime, we found out that there were approximately the same amount of White and Black offenders (44% in 2020). This information meant that the disparity in killings could come from the fact that comparing to their population, there is a bigger proportion of Black people responsible of violent crimes. At first we compared two states to look for a vizual significance but it went totally the other way we thought. In fact the correlation revealed that (A compléter avec les résultats du moedellig). So at this point neither our hypothesis on political party, neither the one on crime were true.

Our final thought is that these killings depend a lot on personal beliefs, behaviors, and variables that cant be measured or explained.

In our first table we were able to compare the circumstances of killings (ou placer ca ?) We observed that 10% of the killings were not explained by the victim presenting a reason to be shot.

  • To conclude, even thought our project didn’t go the way we wanted, we can still confirm that if you’re a Black male in the united states you are in more likely to get killed than a White person male or female.